A photovoltaic inverter operation abnormality detection method based on kernel principal component recombination technology

Abstract

The application discloses a photovoltaic inverter operation abnormality detection method based on a kernel principal component recombination technology, how to further mine directly effective features reflecting abnormal changes by recombination on a plurality of nonlinear principal components extracted by kernel principal component analysis, so as to realize abnormality detection on the operation state of the photovoltaic inverter by monitoring the fluctuation of the features. Specifically, the method of the application generates recombination features that can best reflect the difference by recombining the elements in the kernel principal component vector through maximum-minimum difference feature analysis, and the recombination features are used for photovoltaic inverter operation abnormality detection. Compared with the conventional method of using a nonlinear modeling algorithm to perform abnormality detection, the method of the application obtains the recombination features that can best reflect the difference between online data and normal data in real time through recombination. Therefore, the method of the application has the significant features of self-adaptation and updating change in theory.

Description

Technical Field

[0001] This invention relates to a method for detecting abnormalities in photovoltaic equipment, and more particularly to a method for detecting abnormalities in the operation of a photovoltaic inverter based on core principal component recombination technology. Background Technology

[0002] A photovoltaic (PV) inverter converts the direct current (DC) voltage generated by photovoltaic (PV) solar panels into alternating current (AC) at the mains frequency. PV inverters are a crucial component of PV array systems, typically used in conjunction with AC-powered equipment, and are among the most closely monitored devices in the daily operation and maintenance of PV power plants. However, in actual operation, PV inverters are affected not only by changes in their own electrical signals but also by the surrounding environment (e.g., temperature). The changing characteristics of most PV inverter systems in practical applications are complex and variable, making it impossible to directly establish a precise mechanistic model to describe the different operating states of the PV inverter. Therefore, using a mechanistic model to monitor the operating status of PV inverters is not advisable.

[0003] In fact, since most large-scale photovoltaic power plants in my country are located in the northwest region, where the environment is relatively harsh, inverter maintenance is quite difficult. If an inverter malfunctions but is not detected and addressed in a timely manner, it can lead to irreparable economic losses and safety hazards. Therefore, real-time detection of photovoltaic inverter malfunctions has significant practical value. On the other hand, photovoltaic power plants are already equipped with corresponding Supervisory Control and Data Acquisition (SCADA) systems. These systems can measure and store real-time monitoring data of the photovoltaic inverter and external environment at various sampling times according to a fixed sampling frequency. This data undoubtedly contains useful information for operators to detect any operational abnormalities.

[0004] Furthermore, guided by current industrial big data, the use of photovoltaic inverter monitoring data and environmental monitoring data to drive the detection of photovoltaic inverter operational anomalies has received increasing research and attention. With the ever-increasing demands for the reliability and safety of photovoltaic power generation systems, and considering the complex characteristics of real-time monitoring data from photovoltaic power plants, such as nonlinearity, analyzing real-time monitoring data from photovoltaic power plants, deeply exploring its inherent nonlinear relationships, and extracting directly effective implicit component information for photovoltaic inverter operational anomaly detection is of great significance. While kernel principal component analysis (KPI) is a classic anomaly detection method applicable to nonlinear data, its ability to achieve infinite-dimensional nonlinear implicit mappings makes it impossible to determine which kernel principal component is most directly effective for anomaly detection. Therefore, this type of method still requires further improvement and refinement. Summary of the Invention

[0005] The main technical problem this invention aims to solve is: how to further extract directly effective features that reflect abnormal changes from multiple nonlinear principal components extracted by kernel principal component analysis (KPI), and then detect anomalies in the operating status of photovoltaic inverters by monitoring the fluctuations of these features. Specifically, the method of this invention first performs KPI transformation on the monitoring data of the photovoltaic inverter, then recombines the KPIs through minimization-maximum difference feature analysis to extract the features that best reflect anomalies. Finally, it detects whether the photovoltaic inverter is operating abnormally by monitoring the fluctuations of these features.

[0006] The technical solution adopted by the present invention to solve the above problems is as follows: a method for detecting abnormal operation of a photovoltaic inverter based on core principal component recombination technology, comprising the following steps:

[0007] Step (1): Obtain the photovoltaic inverter's data under normal operating conditions from the historical sampling database corresponding to the photovoltaic inverter. Group sampling data, and record each group of sampling data separately. A dimensional data vector, thus obtaining the corresponding... Data vectors Then, Composition of data matrix Among them, the first Data vectors Specifically by The data consists of monitoring data from a photovoltaic inverter, labeled as follows: , express A 1-dimensional real vector It represents the set of real numbers.

[0008] Step (2): According to For data matrix The row vectors of each row in the matrix are standardized to obtain the reference data matrix. ;in, express The Middle The row vector of a row. express The Middle The row vector of a row. , and Representing row vectors respectively The mean and standard deviation of all elements in the dataset. express A real matrix of dimension 1.

[0009] Step (3): Determine the kernel parameters Then, the kernel principal component transformation matrix is ​​determined according to steps (3.1) to (3.3) as shown below. ;in, express A 3D real matrix It equals the number of kernel principal components.

[0010] Step (3.1): Calculate the kernel matrix according to the formula ① shown below. :

[0011] ①

[0012] in, Representing the kernel matrix The Middle Line number Column elements, , , This represents an exponential function with base e. and Representing the reference data matrix The Middle row and number The column vector of a column. Indicates the computation of column vectors and column vectors The distance between them, where the superscript T indicates the transpose of the matrix or vector.

[0013] Step (3.2): According to the formula Kernel matrix Centralized processing is performed to obtain the reference kernel matrix. ; where the matrix All elements in the array are equal to 1.

[0014] Step (3.3): Determine the number of kernel principal components equal to Then, solve for the reference kernel matrix. The largest The eigenvectors corresponding to each eigenvalue Then Composition of the kernel principal component transformation matrix .

[0015] Step (4): According to the formula The kernel principal component matrix was calculated. Then, in turn, respectively The column vectors in the matrix are treated as kernel principal vectors. and to Perform minimization of difference feature analysis to obtain the recombined features. The specific implementation process is shown in steps (4.1) to (4.5).

[0016] Step (4.1): Initialization Then set Equal to the kernel principal component matrix The first in Column vector.

[0017] Step (4.2): Except for the first The rest besides column vectors The column vectors form the reference kernel principal component matrix. ;in, express A real matrix of dimension 1.

[0018] Step (4.3): Calculation Each column vector in the middle and The distance between them, and then Zhongyu The closest distance between Each column vector is labeled as ;in, equal to a less than Positive integers.

[0019] Step (4.4): According to the formula Calculate the coefficient vector Then, solve the eigenvalue problem. The largest eigenvalue in the middle corresponding feature vector And according to the formula Calculate the recombined vector ;in, .

[0020] Step (4.5): According to the formula Calculate the recombination features Then, make a judgment Is it less than If so, then set Then set Equal to the kernel principal component matrix The first in Column vector and return to step (4.2); otherwise, obtain Recombination features .

[0021] Step (5): Calculation average and standard deviation Then, set the initial change limit for each. and the lower limit of initial change .

[0022] It should be noted that the implementation process of solving the eigenvalue problem in step (4.4) above is actually a... The invention employs a maximum-minimum discrepancy feature analysis technique. This technique considers two optimization objectives: maximizing the variance through recombination of the kernel principal component vectors. The corresponding nearest neighbor reconstruction error and the minimization of the reference kernel principal component matrix The fluctuation changes corresponding to each column vector.

[0023] The corresponding nearest neighbor reconstruction error is achieved by recombining vectors. sum coefficient vector The calculation yields the following result:

[0024] ②

[0025] Minimize the reference kernel principal component matrix The fluctuations corresponding to each column vector are achieved through the recombination of vectors. Implementation, that is:

[0026] ③

[0027] Combining the maximization and minimization problems in ② and ③, the objective function can be constructed as follows:

[0028] ④

[0029] By constructing the Lagrange function This allows us to solve equation ④ above. First, we need to calculate... Compared to Partial differential equation:

[0030] ⑤

[0031] Then, by setting the partial differential to zero, the objective function in equation ④ above takes an extreme value, and the generalized eigenvalue problem can be deduced. Furthermore, due to The row vectors in the vector array are mutually orthogonal. The result of the calculation is equal to a diagonal matrix, so its inverse matrix can be calculated directly. In other words, in the generalized eigenvalue problem, both sides of the equation are multiplied by the left side. This leads to the eigenvalue problem in step (4.4): .

[0032] because The calculation result is A column vector of dimension, The rank of is equal to 1, therefore the eigenvalue problem in step (4.4) has only one non-zero eigenvalue, which is the largest eigenvalue.

[0033] Step (6): Obtain a set of sampling data from the photovoltaic inverter at the latest sampling moment, and record and store it as a single data set. 3D data vector Then, according to right The elements in each row are standardized to obtain a column vector. ;in, and They represent and The first in Row elements, data vector In The order of the monitoring data for each photovoltaic inverter is the same as in step (1).

[0034] Step (7): Calculate the kernel vector according to formula ⑥ shown below. Then, through the formula right Centralized processing is performed to obtain online kernel vectors. :

[0035] ⑥

[0036] in, express The first in Element, subscript row vector All elements in the array are equal to 1.

[0037] Step (8): According to the formula The kernel principal component vector is calculated. Then set Afterwards, Perform a minimization-maximum difference feature analysis to obtain the corresponding recombination vector. The specific implementation process is shown in steps (4.3) to (4.4).

[0038] Step (9): According to the formula Calculate the recombination features Then, according to the formula Calculate the reference recombined feature vector Then The largest and smallest elements are recorded as follows: and .

[0039] Step (10): Determine the upper limit of change and lower limit of change Then, determine whether the conditions are met. If so, then the photovoltaic inverter is operating normally, according to the formula. Update the kernel principal matrix Then, set Return to step (6); if not, then the photovoltaic inverter is malfunctioning; among which, Indicates selection and The maximum value in, Indicates selection and The minimum value in.

[0040] The advantages of the method of the present invention, based on the above implementation steps, are as follows.

[0041] First, the method of this invention further performs maximum-minimum difference feature analysis on the kernel principal component vector, and recombines the elements in the kernel principal component vector to generate recombination features that best reflect the differences, which are used for anomaly detection in photovoltaic inverters. Compared with traditional methods that use nonlinear modeling algorithms for anomaly detection, the method of this invention, through recombination, obtains recombination features in real time that best reflect the differences that may exist between online data and normal data. Therefore, theoretically, the method of this invention has the significant characteristics of adaptability and updating. Attached Figure Description

[0042] Figure 1 This is a schematic diagram illustrating the implementation process of the method of the present invention. Detailed Implementation

[0043] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0044] This invention discloses a method for detecting operational anomalies in photovoltaic inverters based on core principal component recombination technology. The following is a detailed explanation of such methods. Figure 1 The implementation flowchart shown below illustrates the specific implementation method of the present invention.

[0045] Step (1): Obtain the photovoltaic inverter's operating conditions from the historical sampling database of the photovoltaic inverter. Group sampling data, and record each group of sampling data separately. A dimensional data vector, thus obtaining the corresponding... Data vectors Then, Composition of data matrix .

[0046] Specifically, the Individual data vectors In The monitoring data for each photovoltaic inverter are arranged in the following order: DC input current, DC input voltage, AC output current, AC output voltage, photovoltaic inverter operating time, photovoltaic inverter power, photovoltaic inverter temperature, photovoltaic power plant power generation, photovoltaic power plant power, ambient temperature, ambient humidity, radiation, and wind speed.

[0047] Step (2): According to For data matrix The row vectors of each row in the matrix are standardized to obtain the reference data matrix. .

[0048] Step (3): Determine the kernel parameters Then, the kernel principal component transformation matrix is ​​determined according to the aforementioned steps (3.1) to (3.3). The specific implementation process of step (3.3) is shown in steps (i) to (iv).

[0049] Step 1: Solving the eigenvalue problem All The eigenvectors corresponding to each eigenvalue are obtained and sorted in descending order according to the magnitude of the eigenvalues, resulting in the descending order of the eigenvalues. and its corresponding eigenvectors ;in, Represents the eigenvalues ​​in the eigenvalue problem. This represents the eigenvector in the eigenvalue problem.

[0050] Step 2: According to the formula For eigenvectors Normalization is performed; where the subscript number ;in, Indicates the first eigenvalues The corresponding feature vector.

[0051] Step 3: Confirm To meet the conditions After determining the smallest positive integer, the number of kernel principal components is then set to equal to... ;in, Indicates the first 1 feature value, index .

[0052] Step 4: Place the front The largest eigenvalue corresponding feature vector Composition of the kernel principal component transformation matrix .

[0053] Step (4): According to the formula The kernel principal component matrix was calculated. Then, in turn, respectively The column vectors in the matrix are treated as kernel principal vectors. and to Perform minimization of difference feature analysis to obtain the recombined features. The specific implementation process is shown in steps (4.1) to (4.5).

[0054] Step (5): Calculation median average and standard deviation Then, calculate the initial upper limit of change. and the lower limit of initial change .

[0055] Step (6): Obtain a set of sampling data from the photovoltaic inverter at the latest sampling moment and store it as a single data set. 3D data vector Then, according to right The elements in each row are standardized to obtain a column vector. .

[0056] Step (7): Calculate the kernel vector according to formula ⑥ Then, through the formula right Centralized processing is performed to obtain online kernel vectors. .

[0057] Step (8): According to the formula The kernel principal component vector is calculated. Then set Afterwards, Perform maximum-minimum difference feature analysis to obtain the corresponding recombination vector. The specific implementation process is shown in steps (4.3) to (4.4).

[0058] Step (9): According to the formula Calculate the recombination features Then, according to the formula Calculate the reference recombined feature vector Then The largest and smallest elements are recorded as follows: and .

[0059] Step (10): Determine the upper limit of change and lower limit of change Then, determine whether the conditions are met. If so, then the photovoltaic inverter is operating normally, according to the formula. Update the kernel principal matrix Then, set Return to step (6); if not, then the photovoltaic inverter is malfunctioning; among which, Indicates selection and The maximum value in, Indicates selection and The minimum value in.

Claims

1. A method for detecting operational anomalies in photovoltaic inverters based on core principal component recombination technology, characterized in that, Specifically, the steps are as follows: Step (1): Obtain the photovoltaic inverter's operating conditions from the historical sampling database. Groups of sampled data were collected, and each group of sampled data was recorded separately. A dimensional data vector, thus obtaining the corresponding... Data vectors Then, Composition of data matrix Among them, the first Data vectors Specifically by The data consists of monitoring data from a photovoltaic inverter, labeled as follows: , express A 1-dimensional real vector Represents the set of real numbers; Step (2): According to For data matrix The row vectors of each row in the matrix are standardized to obtain the reference data matrix. ;in, express The Middle The row vector of a row. express The Middle The row vector of a row. , and Representing row vectors respectively The mean and standard deviation of all elements in the dataset. express A 3D real matrix; Step (3): Determine the kernel parameters Then, the kernel principal component transformation matrix is ​​determined according to steps (3.1) to (3.3) as shown below. ;in, express A 3D real matrix Equal to the number of kernel principal components; Step (3.1): Calculate the kernel matrix according to the formula ① shown below. : ① in, Representing the kernel matrix The Middle Line number Column elements, , , This represents an exponential function with base e. and Representing the reference data matrix The Middle row and number Column vectors of columns Indicates the computation of column vectors and column vectors The distance between them, where the superscript T indicates the transpose of the matrix or vector; Step (3.2): According to the formula Kernel matrix Centralized processing is performed to obtain the reference kernel matrix. ; where the matrix All elements in the array are equal to 1; Step (3.3): Determine the number of kernel principal components equal to Then, solve for the reference kernel matrix. The largest The eigenvectors corresponding to each eigenvalue Then Composition of the kernel principal component transformation matrix ; Step (4): According to the formula The kernel principal component matrix was calculated. Then, in turn, respectively The column vectors in the matrix are treated as kernel principal vectors. and to Perform minimization of difference feature analysis to obtain the recombined features. The specific implementation process is shown in steps (4.1) to (4.5); Step (4.1): Initialization Then set Equal to the kernel principal component matrix The first in Column vector; Step (4.2): Except for the first The rest besides column vectors The column vectors form the reference kernel principal component matrix. ; Step (4.3): Calculation Each column vector in the middle and The distance between them, and then Zhongyu The closest distance between Each column vector is labeled as ;in, , express A 3D real matrix; Step (4.4): According to the formula Calculate the coefficient vector Then, solve the eigenvalue problem. The largest eigenvalue in the middle corresponding feature vector And according to the formula Calculate the recombined vector ;in, ; Step (4.5): According to the formula Calculate the recombination features Then, make a judgment Is it less than If so, then set Then set Equal to the kernel principal component matrix The first in Column vector and return to step (4.2); otherwise, obtain Recombination features ; Step (5): Calculation average and standard deviation Then, calculate the initial upper limit of change. and the lower limit of initial change ; Step (6): Obtain a set of sampling data from the photovoltaic inverter at the latest sampling moment and record it as a single data set. 3D data vector Then, according to right The elements in each row are standardized to obtain a column vector. ;in, and They represent and The first in row element; Step (7): Calculate the kernel vector according to formula ② Then, through the formula right Centralized processing is performed to obtain online kernel vectors. : ② in, express The first in Element, subscript row vector All elements in the array are equal to 1; Step (8): According to the formula The kernel principal component vector is calculated. Then set Afterwards, Perform a minimization-maximum difference feature analysis to obtain the corresponding recombination vector. The specific implementation process is shown in steps (4.3) to (4.4); Step (9): According to the formula Calculate the recombination features Then, according to the formula Calculate the reference recombined feature vector Then The largest and smallest elements are recorded as follows: and ; Step (10): Determine the upper limit of change and lower limit of change Then, determine whether the conditions are met. If so, then the photovoltaic inverter is operating normally, according to the formula. Update the kernel principal matrix Then, set Return to step (6); if not, then the photovoltaic inverter is malfunctioning; among which, Indicates selection and The maximum value in, Indicates selection and The minimum value in.

2. The method for detecting operational anomalies in a photovoltaic inverter based on core principal component recombination technology according to claim 1, characterized in that, The specific implementation process of step (3.3) is as follows: Step 1: Solving the eigenvalue problem All The eigenvectors corresponding to each eigenvalue are obtained and sorted in descending order according to the magnitude of the eigenvalues, resulting in the descending order of the eigenvalues. and its corresponding eigenvectors ;in, Represents the eigenvalues ​​in the eigenvalue problem. Represents the eigenvectors in the eigenvalue problem; Step 2: According to the formula For eigenvectors Normalization is performed; where the subscript number ;in, Indicates the first eigenvalues The corresponding feature vector; Step 3: Confirm To meet the conditions After determining the smallest positive integer, the number of kernel principal components is then set to equal to... ;in, Indicates the first 1 feature value, index ; Step 4: Place the front The largest eigenvalue corresponding feature vector Composition of the kernel principal component transformation matrix .

Images